On Tue, Feb 10, 2015 at 9:52 AM, Xavier Stevens wrote:
> I'm trying to write a logical decoding plugin and I'm seeing some
> interesting behavior around the startup callback. When I use psql and the
> built-in SQL functions (like pg_logical_slot_peek_changes) to use my module
> I see the startup_cb get called. I have written my own streaming replication
> client and for some reason when I use my client the startup_cb does not get
> called, but I can still see the changes just fine. The reason why the
> startup_cb is particularly important in my case, is I'm trying to set two
> dynamic OIDs for PostGIS geometry and geography types. I've tried setting
> them both as globals and as part of the output_plugin_private data.
>
> Any idea what's wrong here? Did I miss something in the streaming
> replication documentation?
Not sure, I would imagine that your problem is in the manipulation of
the memory contexts related to the decoder plugin and the way your
variables are saved.
The documentation states that startup_cb is called each time a
replication slot is created or when a slot is asked to stream changes,
be it through the psql interface or using the replication protocol:
http://www.postgresql.org/docs/devel/static/logicaldecoding-output-plugin.html#LOGICALDECODING-OUTPUT-PLUGIN-CALLBACKS
Just in case, I have just done a quick test with pg_recvlogical on
latest HEAD of REL9_4_STABLE and the startup_cb gets called:
DEBUG: received replication command: IDENTIFY_SYSTEM
DEBUG: received replication command: START_REPLICATION SLOT "slot" LOGICAL 0/0
LOG: called startup_cb_wrapper
Note that on 9.5 (master HEAD at c619c23) I am seeing an assertion
failure. What is the server version you are using? I would imagine 9.4
but let's be sure.
Regards,
--
Michael